Study on Representative Parameters of Reverse Engineering for Maintenance of Ballasted Tracks

Abstract

Reverse engineering (RE) is a technology used to create three-dimensional (3D) models by scanning structures and can be used to examine the current condition of structures. Applying RE to the maintenance of railroad facilities with a high proportion of safety accidents can be an alternative to increase the efficiency of railroad facilities. However, most tasks while constructing Building Information Modeling (BIM) after 3D scanning and extracting two-dimensional (2D) drawings are still performed manually. In particular, denoising, registration, and 3D modeling based on point clouds are labor-intensive and time-consuming tasks, and their efficiency needs to be enhanced by introducing automation technology. In this study, we selected point clouds-based representative parameters for ballasted tracks of a straight single-line section for automating railroad maintenance. Scan data and a BIM of a ballasted track were compared using the selected representative parameters. In addition, the types of damage to ballasted track requiring maintenance were examined. And a testbed was consisted of ballasted a track was selected, and 3D scanning was performed to obtain point cloud data of a testbed. Then, a BIM model was created by measuring the numerical values corresponding to the representative parameters on the scan data. The feasibility of constructing a railroad maintenance BIM based on representative 3D object detection parameters during RE work on the ballasted track was evaluated.

1. Introduction

With recent developments in digital technology, including research on reverse engineering (RE), studies on constructing building information models (BIM) by acquiring as-built structure shape information through three-dimensional (3D) scanning [1]. RE is a technology used to generate drawings of old buildings for which no drawings are available [2] or to acquire shape information about objects and extract 3D models and two-dimensional (2Dloe) drawings using photogrammetry and light detection and ranging (LiDAR) techniques for reviewing the construction of structures [3]. RE using a 3D laser scanner is faster than the conventional method and has the advantage of acquiring information in a noncontact manner [4]. These features make RE suitable for railroad facilities that have no drawings because they were installed a long time ago or those requiring rapid maintenance. Similar to other types of Social Overhead Capital (SOC) facilities, railroad facilities require timely maintenance. They are complex facilities concentrated in small spaces, with components such as rails, sleepers, ballast, and railway catenary system, and the requirement that trains be operated at specified times. Effective maintenance of railroad facilities is considered critical because it is closely related to the incidence of accidents. In particular, the loss of ballast creates voids around the sleeper on a ballasted track. This, in turn, causes the sleeper to be supported improperly. This phenomenon affects train safety and thereby, emphasizes the importance of maintaining railway facility [5,6].

Research on railroad maintenance automation, such as robotic and autonomous systems (RAS) [7], has been conducted. Recently, the application of object detection has been researched as well [8]. In this study, we conducted basic research to construct a BIM of a ballasted track as a measure to automate railroad maintenance.

In general, considerable investments of manpower, financial resources, and time are required to apply RE to facilities [9]. It involves scanning the target structure with a 3D scanner, building a BIM based on the scan data, or extracting 2D drawings. However, as RE is labor-intensive, time-consuming, and costly, the efficiency of RE should be improved by combining it with automation technology [9]. For this reason, automation studies that integrate RE and BIM technologies have gradually increased in building constructions, and studies on the automatic creation of BIM models and automatic extraction of 2D drawings are being conducted. Based on these studies, it can be predicted that the automation ratio will gradually increase over the entire process of Scan-to-BIM, such as linking BIM libraries, building BIM models, and extracting 2D drawings after 3D scans are conducted in the future.

Recently, in the construction field, research on automating 3D scan data processing is being actively conducted. Generally, point cloud data (PCD) extraction by 3D scanning generates noise due to the refraction of light, distortion, equipment errors. [10]. As such noises and outliers affect the accuracy and density of scanned data, they are removed using various denoising filter algorithms based on point cloud data [11]. Furthermore, the multitemporal scanned point cloud data is performed on registration by ICP (iterative closest points) algorithm and similar registration algorithms [12]. These data pre-processing processes of denoising, registration, and adjusting the number of point clouds are essential on PCDs [10]. However, pre-processing railroad data is a labor-intensive and time-consuming process because it deals with a large amount of highly dense point cloud data, similar to the data for an earthwork site [9]. Therefore, a study on automated analysis of digital maps is being conducted to improve the productivity of pre-processing tasks that had previously been performed manually [13]. Object detection to classify and recognize construction equipment, manpower, and materials, which are the main entities present at earthwork sites, is also being researched by various institutions [14]. In the field of railroad engineering, point clouds-based object detection for tunnels and bridges has also been researched [8,15].

In this study, the rails, ballast, and sleepers of ballasted track were selected as research subjects. The combined processes of object detection and scan-to-BIM, illustrated in Figure 1, were researched assuming that the maintenance of railroad facilities have been automated. It consists of five steps: (1) collection of point cloud data through 3D scanning; (2) point cloud-based object detection; (3) 3D scan data review using representative parameters for rails, ballast, and sleepers during object detection; (4) linking of railroad BIM library elements corresponding to rails, ballast, and sleepers, based on the object detection results; and (5) construction of a railroad BIM for maintenance. The scope of the current study encompassed two of these steps, i.e., (3) 3D scan data review using representative parameters for rails, ballast, and sleepers during object detection and (4) linking of railroad BIM library elements corresponding to rails, ballast, and sleepers based on the object detection results.

This paper is organized as follows: Section 2 describes the methods used in this paper to survey the literature. Section 3 represents selection of the representative parameters on ballasted track. A testbed experiment of the representative parameters on ballasted track are conducted in Section 4. Section 5 concludes this article. As Figure 2 shows, a literature review on RE and object detection was conducted. In addition, acceptance railroad criteria for track irregularities used by the International Union of Railways (UIC), EN, France, Germany, Italy, Austria, US, Japan, and the Republic of Korea were examined. Representative parameters required for damage type identification and 3D object detection of ballasted track were selected based on the review results. A testbed composed of ballast was used to verify the selected representative parameters and acquired point cloud data by 3D scanning, as shown in Figure 2. The values of the representative parameters were measured using the scan data, and a ballasted track BIM model was created. The created ballasted track BIM model and the testbed scanned point cloud data were compared. This compared task has two main key criteria. (1) the representative parameters for the rail, ballast, and sleeper ware identified whether the shape of the ballasted track could be detected. (2) The representative parameters for the rail, ballast, and sleeper ware identified whether the ballasted track damage types described in the literature review could be examined. Finally, based on the testbed results, the shape types of objects that could be distinguished by the representative parameters and the ballasted track damage types that require maintenance were summarized.

2. Related Works

Scanning using a 3D laser scanner for allows rapid work without contacting objects and facilitates the generation of a 3D model with a smooth surface [2,3]. RE was developed based on the advantages offered by a 3D scanner and can output to Scan-to-CAD or Scan-to-BIM. Early research on RE focused on Scan-to-CAD technology, by which 2D drawings are extracted from 3D scans [16]. Recently, with the development of BIM technology, scan-to-BIM technology has advanced, and many studies have been conducted on generating BIM model in the construction industry [1]. Regardless of Scan-to-CAD or Scan-to-BIM, RE is commonly performed as the process of “3D scanning, point clouds pre-processing, and scanned model generation” [3]. Despite its advantages, RE is still limited in terms of being time-consuming, labor-intensive, and expensive for laser scanning, 3D modeling of scan data, and generating CAD and BIM models [3,8].

Various studies RE automation have been conducted to overcome these limitations. Son et al. [17] studied an automated laser scanning system that automatically performs laser scanning and quality inspection using a CNC-type laser scanner. For the automated laser scanning system research, they developed a software application that could automatically perform a 3D scan using a CNC-type laser scanner, scan data registration, X, Y, Z coordinate conversion, and 3D modeling. The accuracy of the automatically constructed 3D scan model was verified by comparing it with a 2D CAD model. Pu and Vosselman [18] conducted a study on the automated generation of a building model by creating a geometric model from the features and boundaries of a building from Terrestrial Laser Scanning (TLS) data. Guan and Gu [19] generated a 3D model with a smooth surface more quickly than possible by conventional methods by 3D scanning the complex hull and propeller of a ship and removing noise using the ICP algorithm.

Scan-to-CAD has been consistently studied as well. Roseline et al. [20] automatically generated a 3D mesh model after laser scanning and a 3D CAD model based on geometric information. Although they automated the generation of a 3D CAD model by combining RE and CAD, their approach is not suitable for application in the maintenance of railroad facilities composed of relatively large materials because it was conducted using cylinders in a laboratory.

With recent development in BIM technology, studies on scan-to-BIM are being actively conducted. In particular, applications of scan-to-BIM to the restoration and maintenance of historical buildings are increasing. Hichri et al. [21] proposed that scan-to-BIM is necessary for efficient restoration, documentation, and maintenance of historic buildings. Furthermore, they applied the manual steps of acquisition of point cloud data, segmentation of material properties, and the construction of a BIM model to the process. Ham et al. [2] proposed the automated registration method of point cloud data by 3D scanning of complex cultural heritages and extracted a CAD file (*.dwg). They also proposed the phased RE framework for generation a BIM model using the AUTODESK Revit software by extracting the IFC (Industry Foundation Classes) library from a geometric model. Although Ham et al. [2] studied the automated registration method of point cloud data extracted by a 3D scanner, they performed the other processes manually and proposed that further research is required to fully automated RE.

Considering the above reviews of previous studies, Scan-to-BIM automatically extracts a CAD file, but more complex steps are required to generate a 3D BIM model. Thus, more research is needed to achieve full automation. As part of the effort to overcome the technological limitations to automation, various studies have been conducted on automated 3D modeling by applying semantic segmentation and object classification based on deep learning algorithms.

Xiong et al. [22] studied semantic segmentation by applying a context-based modeling algorithm. They scanned the indoor space of buildings using TLS and automated the conversion of the raw point cloud data for a building into a BIM model. In addition, the main components of the indoor environment, such as walls, floors, ceilings, windows, and doors, were modeled at the same time, with the windows and doors extracted by a semantic segmentation algorithm. Zeng et al. [23] generated a 3D model from point cloud data acquired by 3D scanning of a building and then detected the components of the building, such as windows, doors, and columns, by applying their developed semi-automated method. They proposed a semi-automated method to detect building components using a pre-trained deep neural network without applying scan-to-CAD and scan-to-BIM methods to raw scanned 3D data. Chao et al. [9] automated the extraction of geometric structure information from 3D buildings and performed downsizing, boundary detection, and building component categorization using the raw data. In addition, they conducted an object recognition experiment on building components such as outer walls, doors, windows, and roofs. Studies on object detection have also been conducted in the railroad field. Lee et al. [8] acquired 3D point-clouds through Lidar scanning for a railroad tunnel, generated a 3D model of the structure, and applied semantic segmentation.

According to the results of Reverse Engineering’s literature review examine in this paper, RE technology is gradually developing from scan-to-CAD to scan-to-BIM. And the automation rate in the RE process is steadily increasing too. As deep learning-based object detection develops, it is analyzed that this technology will be highly useful in the pre-processing and post-processing processes of point cloud data. In the early stages of research on scan-to-BIM and object detection, most studies were conducted with small indoor objects. However, as the technology has developed, the subjects of research have progressed to large outdoor objects, such as autonomous vehicles, buildings, and heavy construction equipment. Therefore, in keeping with the trend of related research, this study targeted railroad maintenance using Scan-to-BIM for the components of ballasted railroad tracks, such as rails, ballast, and sleepers.

3. Representative Parameters for Ballasted Track

3.1. Maintenance Standards for Ballasted Track

Generally, a ballasted track is referred to as “classical track” or “conventional track” and has been in use longer than concrete track [24]. The term “ballasted track” is used in the International Union of Railways (UIC) Leaflet 719 [25]. The core structure of this ballasted track is composed of rails, fastenings, sleepers, the ballast bed, sub-ballast, and subgrade [24], as shown in Figure 3.

Ballasted tracks have the advantages of a higher elasticity and easier track correction than non-ballasted track treatment but require continuous track management, such as regular repair and correction work. This routine track management implies that ballasted tracks have a higher risk of accidents, such as train derailments, compared to non-ballasted tracks, requiring routine maintenance to limit life cycle cost (LCC). Particularly, compared to non-ballasted tracks, ballasted tracks require routine maintenance because of frequently occurring track irregularities, ballast refills, and sleeper irregularities. For this reason, countries such as EN (European Union), France, Germany, Italy, and Austria, the United States, and Japan perform intensive routine maintenance of ballasted tracks, using detailed acceptance criteria for track irregularities. The German rail infrastructure company, DB Netz AG, make to manage the operation with detailed railroad maintenance standards [26]. DB Netz AG’s “Track Maintenance Standard” provides acceptance criteria for track irregularities according to speed by classifying maintenance activities according to longitudinal level adjustment, twist, height adjustment, and lining.

The EURO Code provides completion criteria for track irregularities in new installations and improvements to tracks [27]. The EURO Code present criteria for track gauge, cross-level, track surface, alignment, and twist. The French national state-owned railway company, SNCF, presents track management standards for track lines that consider train operation safety and comfort. The SNCF classifies management criteria considering track gauge, track surface, alignment, cross-level, and flatness and presents different detailed management criteria based on train speed [27,28]. Italy classifies management items by track gauge, alignment, track surface, cross-level, and twist and provides detailed management criteria based on train speed [27]. Austria classifies track management items according to track surface, cross-level, alignment, twist, and track gauge and provides detailed management criteria based on train speed [27]. The American Public Transportation Association in the US provides standards for ballasted track maintenance on “Rail Transit Track Inspection and Maintenance (2017)” [29]. Particularly, the US standards classify track management items according to track gauge, cross-level, track surface, and alignment, like those of other countries. In addition, they use superelevation as a classification item, and their management standards are subdivided in detail according to train speed. For conventional tracks in Japan, track maintenance guidelines are classified according to track gauge, cross-level, track surface, alinement, and flatness, and detailed criteria are presented according to train speed [29]. In Republic of Korea, the Korea National Railway’s “Track Maintenance Guide (2016)” presents maintenance criteria for general railroads [30]. Similar to the standards of other countries, the Track Maintenance Guide presents detailed routine maintenance criteria based on track gauge, cross-level, level adjustment, and alignment.

The Korean maintenance standards present detailed management criteria for ballast refill and sleeper irregularities for ballasted track, in addition to criteria for major management items, such as track gauge, cross-level, level adjustment, and alignment. The management criteria for ballasted track irregularities managed by each country are summarized in

Table 1. Track irregularity management criteria by country.

Country	Track Gauge	Cross Level	Track Surface	Alignment	Others
EN	●	●	●	●	Twist
France	●	●	●	●	Flatness
Germany	-	-	●	●	Twist
Italy	●	●	●	●	Twist
Austria	●	●	●	●	Twist
US	●	●	●	●	Superelevation
Japan	●	●	●	●	Flatness
Republic of Korea	●	●	●	●	Ballast refill, sleeper irregularities

. As this table shows, the common management criteria for track irregularities among the eight countries are track gauge, cross-level, track surface, and alignment. As summarized in Table 1, Section 3.2 examined the applicability of the main management criteria of track irregularities that are commonly managed in eight countries, such as track gauge, cross level, track surface, and alignment, as representative parameters of this study.

3.2. Decision of the Representative Parameter on Ballasted Track

Representative parameters for ballasted tracks were determined based on the shape dimensions of ballasted tracks by summarizing the maintenance types for ballasted track irregularities, using the maintenance standards of the various countries presented above. The selection of representative parameters for ballasted tracks is an initial stage of study for automation of scan-to-BIM in the RE process for railroad maintenance, and representative parameters were selected for single-line ballasted tracks in a straight section. The selected representative parameters for ballasted tracks (i.e., H_1R–H_5L, W₁–W_8L, and L₁–L₄) are shown in a cross-sectional diagram and top views in Figure 4.

In the selection of representative parameters for ballasted track, four management criteria which are track gauge, cross level, track surface, and alignment reviewed through track irregularity standards of eight countries in this study. However, cross level, track surface, and alignment of track irregularity were excluded in this study. As cross level, track surface, and alignment are associated with geometry rotation information (roll, pitch, and yaw), it is difficult to accurately measure the value of 3D scan models. Therefore, representative parameters for the ballasted track were selected track gauge, ballast refill, sleeper irregularities, and sleeper exposure of ballasted track irregularity management criteria that can be verified relatively simply from differences along the X, Y, and Z-axes of 3D scanning models. A detailed representation of the parameters for the ballasted track shown in Figure 4 and Figure 5 is presented below. First, H_1R and H_1L in Figure 4 on the left and right of the ballasted track represent the height from the floor level (F.L) to the surcharge fill, which is the peak of the ballast shoulder. These parameters can measure the existence of loss of ballast through the height of the ballast surcharge fill. The parameters H_2R and H_2L represent the heights of the left and right ballast shoulders and can be used to verify the existence of loss of the ballast surcharge fill, as with H_1R and H_1L. The parameters H_2R and H_2L are representative parameters used to check the ballast refill management criteria. The parameters H_3R and H_3L represent the ballast height, excluding the left and right ballast shoulders, whereas H_4R and H_4L represent the left and right rail heights. The parameters H_5R and H_5L represent the sleeper exposure and the degree of loss of gravels for sleepers on the left and right of the ballast. The sleeper exposure can be measured through these representative parameters. W₁ in Figure 4 represents the total width of the ballasted track. W_2R and W_2L represent the right width and left width of the ballasted track, respectively, excluding the sleeper length. W_3R and W_3L represent the widths from the right and left surcharge fills with ballast to the ballast end, respectively, whereas W_4R and W_4L represent the right and left widths from the surcharge fill to the sleeper end, respectively. W₅ represents the sleeper length. The W₆ parameter represents the track gauge and relates to the regulation for measuring the track gauge 14 mm below the rail head. Furthermore, W_7R and W_7L represent the right and left rail base widths, respectively. W_8R and W_8L represent the right and left rail head widths, respectively. Consequently, the type of rail installed on the ballasted track can be measured by applying the above-mentioned parameters H_4R and H_4L (representing the rail height), W_7R and W_7L (representing the rail base width), and W_8R and W_8L (representing the rail head width).

L₁ in Figure 5 represents the rail length, L₂ represents the sleeper thickness, and L_3R represents the distance between the sleepers on the left and right of the rail. Finally, these representative parameters can measure sleeper irregularities. L₄ represents the ballast length. The total size of the ballasted track can be expressed by the parameters H_1R, H_1L, W₁, and L₄.

As mentioned above, the shape dimensions of the ballasted track used in the Korea National Railway, including the rail, ballast, and concrete sleepers, were referenced by considering testbed experiments in selecting the ballast representative parameters to be considered in this study. The utilization values of representative parameters for object detection were increased by selecting representative parameters corresponding to track gauge, ballast refill, sleeper irregularities, and sleeper exposure as management criteria for track irregularities in this study.

4. Case Study

4.1. Extracting Scan Data on Ballasted Track

For evaluating the representative parameters for the ballasted track selected in this study, point cloud data were scanned by a 3D laser scanner for the ballasted track installed in a testbed at Uiwang-si, Gyeonggi-do, Republic of Korea (Figure 6). The study used a TLS to perform five measurements to minimize the occlusion in the scanned data of the testbed. First, 3D scans were performed by placing the TLS at the test bed’s four corners and then, on its central part. The TLS was located in the testbed’s central part to prevent occlusion caused by shadows forming on the inner part of the rail. The scanned data was then post-processed to generate a scan file composed of point cloud data.

The ballast, rail, and concrete sleeper were clearly distinguished in the acquired point cloud data, as shown in Figure 7. The 3D laser scanner used a FARO X130 with a length of 5.65 m. A partial section of the scanned 32m testbed was used in this experiment. Using only a part of the testbed was related to the assignment of numbers from (1) to (9) to the concrete sleepers, as shown in Figure 7. The representative parameters applied to each sleeper were measured and used to generate a BIM model for the ballasted track. Relative coordinates were used while scanning the testbed using the TLS. Because this study is a conceptual study for new technology applications, 3D scanning was performed without GNSS linking. However, absolute coordinate synchronization would be necessary for future field applications.

In addition, the study required more than 10 h for the testbed measurement and BIM modeling. The process consists of testbed 3D scanning, conversion of scanned file, manual measurement of representative parameters, and BIM modeling. The development and application of an automated scan-to-BIM modeling technology through research would save a significant amount of time compared with the current process.

As shown in Figure 8, one scan data was created by cutting the middle point between (1) and (2) based on the concrete sleeper in the 3D scan data of the testbed, and another scan data were created by cutting the middle point between (2) and (3). In this way, scan data of nine ballasted tracks in total were constructed by cutting the scan data at the middle points of sleepers. From these scan data, the values of the representative parameters were measured as shown in Figure 8.

4.2. Generating the BIM Model

The BIM model of the 3D scanned testbed had different values for the surcharge fill height of the ballast shoulder, the concrete sleeper interval, and the concrete sleeper dimensions from the ballasted track used in the field. The BIM model was generated based on the 3D scanned data of the testbed, as shown in Figure 9. For this BIM model, the Revit software of AUTODESK was used.

4.3. Evaluation Experiment for 3D Scan Data and BIM Model

In the evaluation experiment conducted for this study, the shape expression possibilities for the ballasted track and railroad maintenance damage types were examined using the representative parameters selected by comparing the 3D scanning point cloud data for the testbed with the BIM model (

Table 2. Results of the review of possible shape expressions and damage types based on representative parameters.

Representative Parameters	Description	Shape Expression	Track Irregularities Management Criteria
H1R, H1L	Height from floor level (F.L) to surcharge fill (ballast shoulder peak) (R: right, L: left)	Possible	Ballast refill (reduction of ballast shoulder fill) occurs
H2R, H2L	Ballast shoulder height (R, L)	Possible	Ballast refill (reduction of ballast shoulder fill) occurs
H3R, H3L	ballast height excluding ballast shoulder (R, L)	Possible	-
H4R, H4L	Rail height (R, L)	Possible	-
H5R, H5L	Sleeper exposure (R, L)	Possible	Ballast refill (sleeper exposure) occurs
W1	Total ballast width	Possible	-
W2R, W2L	Ballast width excluding sleeper length (R, L)	Possible	-
W3R, W3L	Width from surcharge fill (ballast shoulder peak) to ballast end (R, L)	Possible	Ballast refill (shoulder width reduction) occurs
W4R, W4L	Width from surcharge fill (ballast shoulder peak) to sleeper end (R, L)	Possible	Ballast refill (shoulder width reduction) occurs
W5	Sleeper length	Possible	-
W6	Track gauge	Possible	Track gauge does not occur
W7R, W7L	Rail base width (R, L)	Possible	-
W8R, W8L	Rail head width (R, L)	Possible	-
L1	Rail length	Possible	-
L2	Sleeper thickness (width)	Possible	-
L3R, L3L	Interval between the sleeper edges (R, L)	Possible	Sleeper irregularities (interval irregularity and right-angle irregularity) occur
L4	Ballast length	Possible	-

). The following five damage types and representative parameters of maintenance were verified in this experiment: “track gauge,” “ballast refill (sleeper exposure),” “ballast refill (reduction of ballast shoulder width),” “ballast refill (reduction of ballast shoulder fill),” and “sleeper irregularities (interval irregularity and right-angle irregularity).” The other representative parameters were reviewed simultaneously. To closely examine the representative parameters, the BIM model and scan data for the testbed, with nine concrete sleepers were installed, were divided into the same nine components, based on the sleeper and the center of a sleeper, as shown in Figure 8.

Based on the results of the testbed verification experiment conducted in this study, for track gauge, the distance of a point 14 mm below the rail head of the sleeper (1) was used, as shown in Figure 10. The track gauge was measured as 1435 mm for the sleeper (1) of the testbed, confirming that no track gauge irregularity occurred.

Second, considering the sleeper exposure in the ballast refill, the sleeper exposure was measured for sleeper (1) using the representative parameters as shown in Figure 11. The measured values of the sleeper exposure parameters were H_5R = 50 mm and H_5L = 13 mm.

Third, considering the representative parameters for the reduction of ballast shoulder width damage type in the ballast refill, the same values of W_2R = W_2L = 2200 mm were obtained for the right and left ballast shoulders for sleeper (1) of the testbed. However, the measurements for W_3R and W_3L were different from each other, as were the measurements for W_4R and W_4L, confirming the occurrence of damage associated with reduction of the ballast shoulder width (Figure 12).

Fourth, as a result of reviewing the representative parameters for reduction of ballast shoulder fill in the ballast refill, the values obtained for nine sleepers representing the right and left ballast fills in the testbed were in the ranges of H_1R = 660 ~ 837 mm and H_1L = 580 ~ 624 mm, confirming the occurrence of reduction of ballast shoulder fill (Figure 12).

Finally, based on the review of the interval irregularity and right-angle irregularity (L_3R, L_3L), which is a damage type related to sleeper irregularity, the measurements of L_3R and L_3L for the right and left of sleepers (1)~(3) were different, as shown in Figure 13, confirming the occurrence of sleeper irregularity.

Representative parameters for damage types other than the major damage types for the ballasted track were also examined. H_3R and H_3L represent the ballast height excluding the ballast shoulder. W₁ represents the total ballast width. W_2R and W_2L represent the right and left ballast widths, respectively, excluding the sleeper length. W₅ represents the sleeper length. L₁ represents the rail length. L₂ represents the sleeper thickness (width). L₄ represents the ballast length. These parameters could be verified in this experiment. In addition, the right and left rail heights were examined using H_4R and H_4L, the right and left rail base widths were examined using W_8R and W_8L, and the right and left rail head widths were examined using W_8R and W_8L. The resulting rail head width, base width, and height were found to be 65 mm, 127 mm, and 153 mm, respectively, confirming the installation of 50 kgN rail (50 N).

5. Conclusions

In this study, point clouds-based representative parameters for ballasted tracks were evaluated to assess their potential for use in automated railroad maintenance. A literature review of RE and object detection was conducted, and acceptance standards for railroad track irregularities of the UIC, EU, France, Germany, Italy, Austria, US, Japan, and Republic of Korea were reviewed. Based on the review results, representative parameters required for examination of management criteria and object detection for track irregularities of ballasted track were selected. In addition, the usability of representative parameters was improved by selecting representative parameters for track gauge, ballast refill, sleeper irregularities, and sleeper exposure, which were selected as management criteria for track irregularities in this study.

In addition, point cloud data were acquired by performing 3D laser scanning of a testbed composed of single-line ballasted track in a straight section to evaluate the selected representative parameters. Next, a ballasted track BIM model was generated by measuring the representative parameters for the ballasted track using the acquired 3D point cloud data. This ballasted track BIM was compared with the testbed scanned data to evaluate whether the shape recognition, track gauge, ballast refill, sleeper irregularities, and sleeper exposure of ballasted track could be examined using the selected representative parameters. Through a testbed experiment, this study reviewed the possibility of automating railroad track maintenance using a BIM model based on the representative parameters identified during the RE work for the ballasted track. The result of this study is expected to be helpful in applied scan-to-BIM research on automation of railroad maintenance in the future. In addition, a comparison of the method in this paper and the existing detection method in future studies is likely to contribute to the development of railway maintenance technology.

However, one limitation of this study is that among the management criteria identified for track irregularities, the track gauge, ballast refill, sleeper irregularities, and sleeper exposure were reflected in the selection of representative parameters for the ballasted track, whereas the cross-level, track surface, and alignment were not reflected. Another limitation is that the evaluation experiment was conducted using a testbed rather than a ballasted track in actual operation. The application of the fully automated Scan-to-BIM technology developed by this study may result in a further improvement in accuracy and the solution of “over correctness” in BIM modeling. For these reasons, it is necessary to continue to study the application of object detection algorithms and automated construction of scan-to-BIM models in the railroad maintenance filed in the future, based on the results of the representative parameters for ballasted track derived in this study.